Big Model Daily, December 14

News6months agoupdate AIWindVane
27 0
Big Model Daily, December 14

[Big Model Daily, December 14] Google DeepMind’s most advanced Imagen 2 is released: Can you identify these AI pictures? ;A picture or text generates an infinite 3D world! The new work of Stanford Wu Jiajun’s team has made netizens call it “unbelievable”; the inference efficiency of large models has been improved by 3 times without loss, and the University of Waterloo, Peking University and other institutions have released EAGLE; there is still competition for the “Queen of News”, AI will broadcast news 24/7

Google DeepMind’s most advanced Imagen 2 released: Can you identify these AI pictures?

Google has been getting really good news lately. First, the big killer Gemini was released a few days ago. The series of demo displays at the press conference were dazzling. In just one week, Google has announced some good news: the first version of Gemini Pro is now accessible through the Gemini API; the release of Imagen 2; and the launch of MedLM, a series of models fine-tuned for the medical industry.

A picture or text generates an infinite 3D world! The new work of Stanford Wu Jiajun’s team made netizens call it “unbelievable”

Stanford Wu Jiajun’s team created the AI version of “Alice in Wonderland” masterpiece! Infinitely coherent 3D scenes can be generated along the camera trajectory with just a picture or a piece of text. This work is called WonderJourney and was jointly created by Stanford Wu Jiajun’s team and Google Research.

The inference efficiency of large models has been improved by 3 times without loss, and the University of Waterloo, Peking University and other institutions released EAGLE

Large language models (LLM) are increasingly used in various fields. However, their text generation process is expensive and slow. This inefficiency is attributed to the algorithm of autoregressive decoding: the generation of each word (token) requires a forward pass, requiring access to an LLM with billions to hundreds of billions of parameters. This results in traditional autoregressive decoding being slower. Recently, the University of Waterloo, the Canadian Vector Institute, Peking University and other institutions jointly released EAGLE, which aims to improve the inference speed of large language models while ensuring a consistent distribution of model output text. This method extrapolates the second top-level feature vector of LLM, which can significantly improve the generation efficiency.

ChatGPT becomes Nature’s top ten figures of the year, the first non-human entity

2023 has come to an end, and it’s time to look back at the year’s important scientific advances. On December 13, the famous scientific journal “Nature” just released the top ten people of 2023 (Nature’s 10), which aims to select individuals who have made important contributions to the scientific field and attracted attention to key issues. What is different from the past is that this year’s Nature’s 10 includes the large language model (LLM) ChatGPT, which is the first time a “computer program” has been selected. “Nature” stated that this approach aims to recognize the role played by artificial intelligence systems that imitate human language in the development and progress of science.

Why are you fighting for the “Queen of News”? AI will broadcast news 24/7

“I’m going to be prime-time news seven days a week,” is a line from the recent hit TV series “News Queen.” The play tells the story of several anchors fighting openly and secretly within the TV station. Everyone is working hard to broadcast the prime time for seven consecutive days throughout the year. The audience couldn’t help but sigh that they really love going to work. However, as human beings, no matter how much they love to work, human anchors will eventually have to rest. At this point, the advantages of AI anchors are obvious. They can broadcast news 24/7, just like this demo released by the American media startup Channel 1. The release of this demo sparked heated discussions in the American press because none of the anchors featured in it were real people. Their images are lifelike, their voices are smooth, and even their lip movements are perfectly matched. Moreover, they can switch languages in one second and broadcast news in various languages, which is something that most human anchors cannot do.

Elad Gil leads Harvey’s new funding round at $700 million valuation, along with Sequoia America and OpenAI

Harvey is an AI-powered legal software startup founded in 2022 by former Meta AI researcher Gabe Pereyra and former O’Melveny & Myers attorney Winston Weinberg, aiming to relieve lawyers of the enormous workload of conducting research and due diligence. quantity. According to Information reports, Harvey is undergoing financing negotiations and is expected to raise US$70 to US$80 million, with a valuation of US$700 million, an increase of more than 4 times from the valuation announced in April. The person leading this round of financing is venture capital. Companies Kleiner Perkins and Elad Gil, other investors include Sequoia Capital and OpenAI Startup Fund.

Training efficiency increases 40 times! Open source pictures generate 3D models, Stable Zero123 is here

On December 14, the famous generative AI open source platform was open sourced on the official website. Pictures generate high-quality 3D models – Stable Zero123. Stable Zero123 is developed based on the Zero123 model jointly open sourced by Toyota Research Institute and Columbia University in March this year. The model has been greatly optimized mainly by changing the rendering data set and fractional distillation. Not only does the generated 3D model perform better than Zero123, but the training efficiency is also increased by 40 times. It is worth mentioning that Stable Zero123 can be used in conjunction with’s latest open source high-precision image model SDXL, which is equivalent to a 3D model extension plug-in.

OpenAI: Launches Converge 2 accelerator program, participants will receive $1 million in venture capital

OpenAI teamed up with Microsoft and other investors to establish the OpenAI Startup Fund with a scale of US$100 million, aiming to promote the sustainable development of dozens of AI start-ups in medical, education, culture and other industries. In December 2022, OpenAI will further launch the “Accelerator” (Converge) program. Today, OpenAI founder Sam Altman said that the OpenAI Venture Fund has launched the Converge 2 accelerator program.

Mozilla launches Solo: using AI to help users create websites with zero programming

Mozilla recently launched a new project called Solo, which is aimed at users without any programming experience. By incorporating AI capabilities, the websites created can be comparable to the development effects of professional developers.

Lightning AI launches AI development and deployment platform Lightning AI Studio

Lightning AI launches AI development and deployment platform Lightning AI Studio

Google launches Duet Al for Developers programming assistance tool, which will be equipped with Gemini model in the future

Duet Al for Developers, Google’s suite of AI-assisted tools for code completion and generation announced earlier this year, is now generally available today and will begin using Google’s more powerful Gemini model in the coming weeks.

Gemini is now available for developers to access for free via Google Al Studio

Gemini Pro and Gemini Pro Vision are now available for free through Google Al Studio (formerly MakerSuite). Aimed at individual developers and teams, it is a web-based tool that allows developers to develop tests and prompts for using models, then provides application programming interface keys that can be exported to an application or other development environment.

© Copyright notes

Related posts

No comments

No comments...